Factory AI Droid

ReviewAudited by ClawScan on May 10, 2026.

Overview

The skill is not clearly malicious, but it hands a third-party coding agent broad power to change and deploy code, add extensions, and retain org-wide context without enough provenance or guardrails.

Before installing, verify the Factory droid binary and any plugins/MCP servers, use least-privileged credentials, avoid --force unless you explicitly asked for it, and require review before commits, PRs, or deployments. Also check Factory's session-memory and code-indexing controls for private repositories.

Findings (5)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

The agent could make code changes or trigger deployment workflows in ways that are hard to review or reverse if used too broadly.

Why it was flagged

These commands can mutate repositories, create release/deployment effects, and bypass confirmation; the skill does not define approval or containment requirements before those actions.

Skill content
droid exec "commit my changes with a good message"
droid exec "deploy to fly.io"
droid exec --force "fix lint errors"    # Auto-apply without confirmation
Recommendation

Use this only with explicit user approval for commits, PRs, deployments, and any --force mode; prefer dry-runs, branches, and manual review before applying changes.

What this means

Untrusted or misconfigured CLI/plugins/MCP servers could gain access to code, credentials, or deployment workflows.

Why it was flagged

The skill depends on a preinstalled external binary and can add MCP servers/plugins, but the artifacts provide no source, verification, permissions model, or install provenance.

Skill content
Already installed at: `/Users/mitchellbernstein/.local/bin/droid`
...
droid mcp add server-name         # Add MCP server
droid plugin add name             # Add plugin
Recommendation

Verify the droid binary source and version, install only trusted plugins/MCP servers, and review extension permissions before use.

What this means

The droid CLI may act with the permissions of the logged-in Factory account or API key.

Why it was flagged

Factory authentication is expected for this integration, but it gives the CLI account-level authority; the artifacts do not show hardcoded keys or credential leakage.

Skill content
droid login
# or set FACTORY_API_KEY env var
export FACTORY_API_KEY=your-api-key
Recommendation

Use least-privileged credentials where possible and avoid exposing API keys in shared shells, logs, or project files.

What this means

Private code or prior-session instructions could be reused unexpectedly or influence future tasks.

Why it was flagged

The skill advertises broad organizational code context and persistent session memory without explaining path scope, retention, reuse across tasks, or controls against stale/poisoned context.

Skill content
- Droid has deep codebase understanding across your org
- Session-based memory for context continuity
Recommendation

Confirm what Factory stores, how to clear sessions, and whether org-wide code access can be limited to the repositories needed for the current task.

What this means

Added MCP servers may see prompts, code context, or tool calls depending on their permissions.

Why it was flagged

MCP support is disclosed and purpose-aligned, but the skill does not describe server identity, trust boundaries, or what data/tools added servers can access.

Skill content
MCP servers for extended capabilities
...
droid mcp add server-name         # Add MCP server
Recommendation

Only connect trusted MCP servers and review their access scope before letting the agent use them.